Improving trends estimates for specific work-related ill-health conditions when excess zeros are present in a voluntary health reporting scheme

• Published in: Occupational and environmental medicine (London, England) England), 2023-05, Vol.80 (5), p.280-286
• Main Authors: Gittins, Matthew, Carder, Melanie, Seed, Martin J, Iskandar, Ireny, Daniels, Sarah Ann Michelle, van Tongeren, Martie
• Format: Article
• Language: English
• Subjects: Asthma; Binomial distribution; Contact dermatitis; Dermatitis; Dermatitis, Contact; Epidemiology; Estimates; Health surveillance; Humans; Incidence; Methodology; Models, Statistical; Musculoskeletal System; Occupational diseases; Occupational Diseases - epidemiology; Occupational Health; Random variables; Respiratory diseases; Skin diseases; Trends; Uncertainty
• ISSN: 1351-0711; 1470-7926
• DOI: 10.1136/oemed-2022-108627

BackgroundTrends in occupational disease incidence are estimated in voluntary reporting schemes such as The Health and Occupational Reporting (THOR) Network in the UK. Voluntary reporting schemes request responses even if no cases are observed to reduce uncertainty in non-response. This may result in false zeros that bias trends estimates. Analysis using zero-inflated models is unsuitable for specific health outcomes due to overestimates of the excess zeros. Here, we attempt to account for excess zeros while investigating condition-specific trends.MethodsZero-inflated negative binomial (ZINB) models were fitted to three THOR work-related ill health surveillance schemes Occupational Skin Disease Surveillance (437 reporters between 1996 and 2019), Occupational Physicians Reporting Activity (1094 between 1996 and 2019) and Surveillance of Work-Related and Occupational Respiratory Disease (878 between 1999 and 2019). The probability associated with a response being a false zero was estimated and applied in weighted negative binomial (wgt-NB) models fitted to specific ill-heath conditions. Three ill-health conditions from the three THOR schemes were considered; contact dermatitis, musculoskeletal and asthma, respectively.ResultsWgt-NB models approximately estimated the incidence rate ratios reported by the ZINB models (eg, EPIDERM; ZINB=0.969, NB=0.963, wgt-NB=0.968) for all health outcome annual trends. This was consistent for specific health outcomes which also tended towards the null (eg, contact dermatitis; NB=0.964, wgt-NB=0.969), indicating potentially overestimated downward trends. Though as the ratio of excess zeros to true zeros decreased in rarer health outcomes, the influence on trends also decreased.ConclusionsThrough weighting, we were able to adjust for excess zeros in health outcome-specific trends estimates. Though uncertainty is still present in underlying reporter behaviour meaning caution should be applied with interpretation of any results.